# Switching from OpenAI or Anthropic

## From OpenAI

Change the base URL, the key and the model name. Nothing else.

```python
from openai import OpenAI

# Before
client = OpenAI(api_key="sk-...")

# After
client = OpenAI(base_url="https://api.continuumlabs.ai/v1", api_key="cl-your-key")

reply = client.chat.completions.create(model="glm-5.2", messages=[{"role": "user", "content": "Hello"}])
```

If your code reads the `OPENAI_BASE_URL` and `OPENAI_API_KEY` environment variables, you can switch without changing code at all.

```bash
export OPENAI_BASE_URL=https://api.continuumlabs.ai/v1
export OPENAI_API_KEY=cl-your-key
```

### What is different

- **Model names.** Use one of [our models](/docs/models). OpenAI model names such as `gpt-4o` are not available.
- **Endpoints.** We support chat completions, embeddings and listing models. The Responses API, Assistants, images and audio endpoints are not available yet.
- **Parameters.** Common parameters work: `temperature`, `max_tokens`, `top_p`, `stop`, `tools`, `tool_choice`, `response_format` and `stream`. Support for less common ones depends on the model.

## From Anthropic

Anthropic's API uses a different request format. Switch to the OpenAI SDK, which takes a few lines to change.

| Anthropic | Continuum |
|---|---|
| `client.messages.create(...)` | `client.chat.completions.create(...)` |
| `system="..."` as its own field | A message with `"role": "system"` first in `messages` |
| `max_tokens` (required) | `max_tokens` (optional) |
| `response.content[0].text` | `response.choices[0].message.content` |
| `tools` with `input_schema` | `tools` with `"type": "function"` and `parameters` |
| `stop_reason` | `choices[0].finish_reason` |

```python
# Before, with Anthropic
message = anthropic.messages.create(
    model="claude-...", max_tokens=1024, system="You are concise.",
    messages=[{"role": "user", "content": "Hello"}],
)
text = message.content[0].text

# After, with Continuum
reply = client.chat.completions.create(
    model="glm-5.2",
    messages=[{"role": "system", "content": "You are concise."}, {"role": "user", "content": "Hello"}],
)
text = reply.choices[0].message.content
```

## Choosing a model

Open models differ in what they are best at. Try your own prompts in the [playground](/playground) before you switch, and see the [scorecard](https://github.com/continuumlabs-ai/cookbook/tree/main/scorecard) for how each model does on real Australian tasks.